Linux: How to monitor the nproc limit

Linux: How to monitor the nproc limit

You probably know about 'nproc' limits in Linux which are set in /etc/limits.conf and checked with 'ulimit -u'. But do you know how to handle the monitoring and be alerted when you're close the fixed limit?

Nproc and ps
Nproc is defined at OS level to limit the number of processes per user. Oracle 11.2.0.4 documentation recommends the following:

oracle soft nproc 2047
oracle hard nproc 16384
 
But that is often too low, especially when you have the Enterprise Manager agent or other java programs running.

Do you want to check that you are far from the limit? then you can use 'ps'. But beware, 'ps' by default does not show all processes.
 
In Linux, when doing multithreading, each thread is implemented as a light-weight process (LWP). And you must use the '-L' to see all of them.

Let's take an example. I have a system where 'ps -u oracle' returns 243 lines. But including LWPs shows a lot more processes which is near the limit:

$ ps h -Led -o user | sort | uniq -c | sort -n
      1 dbus
      1 ntp
      1 rpc
      1 rpcuser
      2 avahi
      2 haldaemon
      2 postfix
    166 grid
    400 root
   1370 oracle
 
So the 'oracle' user has 1370 processes. That's high. And this is the actual number where the nproc limit applies.

'ps -Lf' can show the detail. And even without '-L' we can display the NLWP which is the number of threads per process:

ps -o nlwp,pid,lwp,args -u oracle | sort -n
NLWP   PID   LWP COMMAND
   1  8444  8444 oracleOPRODP3 (LOCAL=NO)
   1  9397  9397 oracleOPRODP3 (LOCAL=NO)
   1  9542  9542 oracleOPRODP3 (LOCAL=NO)
   1  9803  9803 /u00/app/oracle/product/agent12c/core/12.1.0.3.0/perl/bin/perl /u00/app/oracle/product/agent12c/core/12.1.0.3.0/bin/emwd.pl agent /u00/app/oracle/product/agent12c/agent_inst/sysman/log/emagent.nohup
  19 11966 11966 /u00/app/11.2.0/grid/bin/oraagent.bin
1114  9963  9963 /u00/app/oracle/product/agent12c/core/12.1.0.3.0/jdk/bin/java ... emagentSDK.jar oracle.sysman.gcagent.tmmain.TMMain
 
The Oracle 12c EM agent has started 1114 threads and the grid infrastructure 'oraagent.bin' has 19 threads. In addition to that I've a lot of other monothreaded processes. This is how we reach 1370 which is the exact value to compare to the nproc limit.

So what are the good values to set? About the high number of threads for EM agent 12c, there are a few bugs. And I suspect that 1000 threads is too much, especially when checking them with 'jstack' I see that they are "CRSeOns" threads that should not be used in 11.2.0.2 and higher. But that's another problem which I'm currently investigating. When you reach the nproc limit, the user will not be able to create new processes. clone() calls will return EAGAIN and that is reported by Oracle as:

ORA-27300: OS system dependent operation:fork failed with status: 11
ORA-27301: OS failure message: Resource temporarily unavailable
And that is clearly bad when it concerns an +ASM instance or archiver processes.

The goal of the nproc limit is only to prevent 'fork bombs' where a process forks forever and exhausts all resources. So there is no problem to increase this limit. However if you set it high for some users ('oracle' and 'grid' usually), it can be a good idea to monitor the number of processes with the ps h -L above. Because having too many processes is suspect and increasing the limit just hides a process leak and defer the failure.

In 'ps h -L -o' The argument 'h' is there to remove the header line, and '-L' to show all processes including LWP. Then you can count with 'wc -l'.

The proof
In order to be sure that 'ps h -L' gives the exact number, I have tested it. In case you want to check this on your system, here is how to do it. And please report any difference.

First, set your limit to 1024 processes. This is a limit for my user, and the limit is set for my shell and all its child processes:

[oracle@VM211 ocm]$ ulimit -u 1024
Now you can check it:

[oracle@VM211 ocm]$ ulimit -a
core file size          (blocks, -c) 0
data seg size           (kbytes, -d) unlimited
scheduling priority             (-e) 0
file size               (blocks, -f) unlimited
pending signals                 (-i) 15919
max locked memory       (kbytes, -l) 64
max memory size         (kbytes, -m) unlimited
open files                      (-n) 1024
pipe size            (512 bytes, -p) 8
POSIX message queues     (bytes, -q) 819200
real-time priority              (-r) 0
stack size              (kbytes, -s) 10240
cpu time               (seconds, -t) unlimited
max user processes (-u) 1024
virtual memory          (kbytes, -v) unlimited
file locks                      (-x) unlimited

[root@VM211 ocm]# ps h -Lu oracle | wc -l
1023
 
Recommended values for Oracle
Currently this is what is set on Oracle linux 6 for 11gR2 by the preinstall package (in /etc/security/limits.conf):

oracle   soft   nproc    16384
oracle   hard   nproc    16384

For 12c, these are set in /etc/security/limits.d/oracle-rdbms-server-12cR1-preinstall.conf which overrides /etc/security/limits.conf:

oracle soft nproc 16384
oracle hard nproc 16384
And just for your information, here is what is set in the ODA X4-2:

oracle soft nproc 131072

So what do you want to set? You probably don't want it too low and experience 'resource temporarily unavailable'. But what you don't want either is 100000 processes on your server. So my recommendation is to set it high but monitor it when the number of processes reaches something that is not sensible. Then you prevent having the system down in case of process leak, but you can detect it and ask for a patch.

Special thanks to Mr. Franck Pachot for such easy illustration.
Regards
Shahzad